Figure 10.

The whole procedure of feature extraction. Nonnegative matrix factorization (NMF) is used for part-based sequence representation. The template sequence of length n is aligned to the sequences of positive (solid line) and negative (dot line) examples by profile–profile alignment method. Next, each alignment is transformed to (n + 2) -dimensional feature vector that is composed of the alignment scores at n positions, the total alignment score and sequence length. Finally, NMF is applied to feature vector for all except two features: total alignment score and sequence length. These extracted feature vectors are used to train SVM for a target template.